I now mostly use the Instacart UI to review shopping carts that Muse has filled for me. I'd like to be able to speak my feedback to muse while I look at the contents, watching them update as muse makes changes and occasionally tweaking by hand when it’s easier. I want UIs like this for everything: surfaces designed from scratch for collaborating with an agent to get things done together without giving up control over the output. The interesting question here is who will build this. Presumably there is someone at Instacart working on an Instacart Agent right now, but I don't want to juggle a bunch of purpose-built specific-thing-agents. I want one assistant that can do everything for me. It seems obvious that this form factor will win. Meta could probably build a great UI for grocery shopping with Muse, but do they actually want to rebuild all of Instacart's business? Eventually, maybe, but in the meantime they'll face competitive pressure from other assistants that do a better job of integrating with Instacart. This suggests there's some optimal partnership boundary between companies building brains, like meta, and companies building hands, like Instacart. On one side, the brains with the largest number of hands will win, and the other, the hands that work with the largest number of brains will win. If this is right, it seems like a dynamic that will favor incumbents on both sides on of the brain/hand divide.